PostHog Review

1M free events/month, then $0.00005/event. That pricing won't scare a startup and won't embarrass you at Series C either. The data warehouse pulling in Stripe and HubSpot alongside PostHog events is the feature that separates this from Mixpanel or Amplitude — those tools can't do that without a third vendor.

The tradeoff worth naming: one platform for everything means one throat to choke if something breaks. A team running LaunchDarkly plus FullStory has independent fallbacks. PostHog consolidated means a bad deploy touches flags, replay, and analytics at once. That's a real risk to think through before you standardize.

The LLM observability layer — tracking token costs across OpenAI and Anthropic — is early but timely. MIT license means self-hosted is a real option if data residency is on the board's agenda. Pilot it with one product team before the company standardizes.

The data model here is what matters. When funnel data, experiment assignments, session recordings, and error events all share the same underlying event store, you stop stitching and start querying. That's not a demo talking point — that's a schema decision that pays compounding dividends. Competitors like Mixpanel and Amplitude give you excellent analytics but hand you a data silo; PostHog gives you a platform.

The built-in data warehouse with SQL access to Stripe, HubSpot, Postgres, and S3 alongside PostHog events is the sharpest edge here. At $0.00005 per event after 1M free, the cost curve stays manageable until serious scale. The tradeoff: 1 project on the free tier and 6 on paid means multi-product orgs or agencies will feel constrained before they hit the billing ceiling.

If we self-host the MIT-licensed build, we own the data pipeline end-to-end — that's real leverage. If we stay on cloud, the 7-year retention on paid plans and CDP exports to BigQuery or Snowflake keep us from being trapped. Three years in, this is either your entire product data stack or the hub everything else routes through.

Three tiers visible on the pricing page. Free tier: 1M analytics events, 5K session replays, 1M feature flag requests — monthly, no card required. Pay-as-you-go kicks in after those limits: $0.00005/event, $0.005/replay recording. 50-person team burning 10M events/month pays roughly $450/month on analytics alone. Year 3 at 30% event growth lands near $9K annually. Compare that to Mixpanel's per-seat model, which hits $12K+ at similar scale.

The platform consolidates what would otherwise be five vendors: Amplitude, LaunchDarkly, FullStory, Optimizely, and a CDP. Each point solution carries its own contract and SSO add-on. PostHog bundles them with no published SSO tax. Self-hosting (MIT license) removes the bill entirely — procurement cost shifts to infra and ops.

Tradeoff: usage-based billing is unpredictable without discipline. Survey responses at $0.10/each add up fast if triggers aren't scoped tightly. Per-product billing limits exist — use them. No public data on auto-renewal window or termination terms.

The single-data-model architecture is the real story here. When funnel drop-off connects directly to session replays without an export-import hop, that's hours recovered per week. Amplitude and Mixpanel both stop at analytics — you're stitching FullStory or Hotjar on top, then LaunchDarkly for flags, then wondering why your flag state doesn't match your event cohort. PostHog eliminates that join problem at the schema level.

Day-three friction will likely surface around the data warehouse and CDP layer. 120+ integrations sounds like a win until you're debugging a Salesforce sync at 11pm. The docs appear practitioner-written — SDK-first, framework auto-detection, SQL query examples — but the breadth of features means the learning surface is wide. One project on the free tier is a real constraint for agencies or teams running multiple products.

The LLM observability feature — tracking token costs across OpenAI, Anthropic, and LangChain — is genuinely differentiated right now. At $0.00005 per event post-free-tier, the cost model stays reasonable at scale. Tradeoff: self-hosted deployments shift ops burden entirely to your team, and the free tier's 1-year data retention caps longitudinal cohort analysis.

The pitch is consolidation, and PostHog mostly earns it. One million analytics events free per month, 5,000 session replay recordings, a million feature flag requests — these aren't teaser limits. A small team could run real product work for months without paying. Compared to stitching together Mixpanel, LaunchDarkly, and FullStory, the all-same-data architecture means you're correlating session replays with funnel drops without export gymnastics. That's the actual daily win.

The learning curve is real, though. This is a developer-first tool. The setup wizard and auto-detecting SDKs soften day one, but by week two you're writing custom SQL against the data warehouse and configuring flag evaluation modes. Non-engineers will feel the weight. The docs indicate self-hosting is MIT-licensed, which is genuinely rare — but self-hosting also means you own the maintenance.

Mobile is listed as a supported platform, but session replay and the richer debugging surfaces feel built for browser-first. The LLM analytics layer — tracking OpenAI and Anthropic token costs — is the freshest thing here. That's a real gap nobody else fills cleanly yet.

Three green flags upfront. One: the feature list is specific — LLM token cost tracking, local flag evaluation, posthog-js session replay with no extra code. Not vaporware, not padded with synonyms. Two: founded 2020, MIT-licensed, self-hostable. That's an exit valve most SaaS vendors won't give you. Three: the pricing page lists exact per-unit costs. That's rarer than it should be.

Two yellow flags. The platform breadth is real, but breadth is also how tools die — shipping 12 products is hard, and depth gaps vs. dedicated vendors like LaunchDarkly on flag targeting or FullStory on replay are plausible, if not confirmed by this evidence. Also: no public funding data visible here. Could go either way.

Honest read: this is the consolidation play that actually makes sense for a product-engineer-run startup. The open-source core is the moat Amplitude and Mixpanel can't copy easily.